lukfre commited on
Commit
57b3460
·
verified ·
1 Parent(s): 6b8e2c4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +17 -13
README.md CHANGED
@@ -545,35 +545,39 @@ This repo contains several multilingual parallel benchmarks covering European la
545
 
546
  ## Collection
547
  As of **October, 2026**, it includes
548
- - MMLU (our `mmlu-mix` build, cleaned and refined)
549
  - `cais/mmlu`
550
  - `openai/mmmlu`
551
  - `CohereLabs/Global-MMLU`
552
- - `edinburgh-dawg/mmlu-redux`
553
  - `collections/INSAIT-Institute/multilingual-benchmarks`
554
  - `Eurolingua/mmlux`
555
- - BELEBELE (copied from `facebook/belebele`)
556
- - ARC-Easy / ARC_Challenge (`allenai/ai2_arc` + `Eurolingua/arcx`)
557
- - PolyMath (`Qwen/PolyMath`)
558
-
559
-
560
- Sources: lukfre/mmlu-mix (MMLU-Redux-EU), , allenai/ai2_arc + Eurolingua/arcx, Qwen/PolyMath; licences as in each source.
561
 
562
 
563
  ## Schema
564
  Every multiple-choice benchmark gets the same schema, so one set of `lm-eval` templates covers all:
565
 
566
- ```
567
- id item key shared by all languages
568
- (mmlu "subject/idx", belebele "link#qnum", ARC id)
569
  question question text (the passage is separate, in `passage`, for belebele)
570
  choices option texts, already in the NEW order
571
  answer index of the correct option in `choices` (NEW order)
572
  answer_original index of the correct option in the source order
573
  perm perm[new_position] = source position, so choices[i] == source_choices[perm[i]]
574
  shuffled False when the item kept its source order (see "Pinned options")
575
- lang, translation_tier, source_dataset
576
- + mmlu: subject, subject_category, idx; belebele: passage
 
 
 
 
 
 
 
577
  ```
578
 
579
  ## Shuffle rules
 
545
 
546
  ## Collection
547
  As of **October, 2026**, it includes
548
+ - MMLU-mix, our own build, cleaned and refined from:
549
  - `cais/mmlu`
550
  - `openai/mmmlu`
551
  - `CohereLabs/Global-MMLU`
552
+ - `edinburgh-dawg/mmlu-redux-2`
553
  - `collections/INSAIT-Institute/multilingual-benchmarks`
554
  - `Eurolingua/mmlux`
555
+ - BELEBELE, copied from `facebook/belebele`
556
+ - ARC-Easy / ARC_Challenge, copied from `allenai/ai2_arc` + `Eurolingua/arcx`
557
+ - PolyMath, copied from `Qwen/PolyMath`
 
 
 
558
 
559
 
560
  ## Schema
561
  Every multiple-choice benchmark gets the same schema, so one set of `lm-eval` templates covers all:
562
 
563
+ ```output
564
+ <data>
565
+ id item key shared by all languages (mmlu "subject/idx", belebele "link#qnum", ARC id)
566
  question question text (the passage is separate, in `passage`, for belebele)
567
  choices option texts, already in the NEW order
568
  answer index of the correct option in `choices` (NEW order)
569
  answer_original index of the correct option in the source order
570
  perm perm[new_position] = source position, so choices[i] == source_choices[perm[i]]
571
  shuffled False when the item kept its source order (see "Pinned options")
572
+
573
+ <metadata>
574
+ lang
575
+ translation_tier
576
+ source_dataset
577
+
578
+ <custom>
579
+ + mmlu: subject, subject_category, idx
580
+ + belebele: passage
581
  ```
582
 
583
  ## Shuffle rules